제목 색상 유사도 기반 신호등의 배면판 영역 추정 기법( A Method of Estimating the Backboard Area of the Traffic
Light Based on the Color Similarity)
저자명

박호범, 한상훈, 김정후, 한영준

Abstract The traffic light recognition is an essential part of ADAS(Advanced Driving Assist
System) in urban environment. Most of the researches used the color of the traffic light bulb and
the shape feature. However, it is hard to get a consistent color data because of the various light
source and weather. Moreover, the backboard of traffic light also tends to lose their shape
feature when it is exposed by strong sunlight. This paper proposes a method of estimating a
backboard area of the traffic light based on color similarity information, which can be calculated
from the difference image between a RGB image and a normalized RGB image. The traffic light
recognition algorithm is consisted of four steps. It detects the turned on light bulbs using HSV
color space and the filtered miss-detected candidates using shape feature of bulbs. And then it
conducts the estimation of backboard area using the color similarity. Finally, it performs the
template matching on estimated areas for the verification and recognition of the traffic light. As
a result, the proposed method showed 94.4% accuracy and took 0.043 second at each frame.
원문 수록처

 2016년 대한임베디드공학회 추계학술대회(2016.11.10~2016.11.12)  pp81-84

자료유형 국내학술회의